; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefix=GFX9 %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -enable-var-scope -check-prefix=GFX11 %s

; We've separated this file from call-args-inreg.ll since GlobalISel does not support the bfloat type.
; Ideally, we should merge the two files once that support lands.

declare hidden void @external_void_func_bf16_inreg(bfloat inreg) #0
declare hidden void @external_void_func_v2bf16_inreg(<2 x bfloat> inreg) #0

define void @test_call_external_void_func_bf16_inreg(bfloat inreg %arg) #0 {
; GFX9-LABEL: test_call_external_void_func_bf16_inreg:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    s_mov_b32 s17, s33
; GFX9-NEXT:    s_mov_b32 s33, s32
; GFX9-NEXT:    s_or_saveexec_b64 s[18:19], -1
; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
; GFX9-NEXT:    s_mov_b64 exec, s[18:19]
; GFX9-NEXT:    v_writelane_b32 v40, s17, 2
; GFX9-NEXT:    s_addk_i32 s32, 0x400
; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
; GFX9-NEXT:    s_getpc_b64 s[18:19]
; GFX9-NEXT:    s_add_u32 s18, s18, external_void_func_bf16_inreg@rel32@lo+4
; GFX9-NEXT:    s_addc_u32 s19, s19, external_void_func_bf16_inreg@rel32@hi+12
; GFX9-NEXT:    s_mov_b32 s0, s16
; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
; GFX9-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; GFX9-NEXT:    v_readlane_b32 s31, v40, 1
; GFX9-NEXT:    v_readlane_b32 s30, v40, 0
; GFX9-NEXT:    s_mov_b32 s32, s33
; GFX9-NEXT:    v_readlane_b32 s4, v40, 2
; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
; GFX9-NEXT:    s_mov_b32 s33, s4
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: test_call_external_void_func_bf16_inreg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s1, s33
; GFX11-NEXT:    s_mov_b32 s33, s32
; GFX11-NEXT:    s_or_saveexec_b32 s2, -1
; GFX11-NEXT:    scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
; GFX11-NEXT:    s_mov_b32 exec_lo, s2
; GFX11-NEXT:    v_writelane_b32 v40, s1, 2
; GFX11-NEXT:    s_add_i32 s32, s32, 16
; GFX11-NEXT:    s_getpc_b64 s[2:3]
; GFX11-NEXT:    s_add_u32 s2, s2, external_void_func_bf16_inreg@rel32@lo+4
; GFX11-NEXT:    s_addc_u32 s3, s3, external_void_func_bf16_inreg@rel32@hi+12
; GFX11-NEXT:    v_writelane_b32 v40, s30, 0
; GFX11-NEXT:    v_writelane_b32 v40, s31, 1
; GFX11-NEXT:    s_swappc_b64 s[30:31], s[2:3]
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_readlane_b32 s31, v40, 1
; GFX11-NEXT:    v_readlane_b32 s30, v40, 0
; GFX11-NEXT:    s_mov_b32 s32, s33
; GFX11-NEXT:    v_readlane_b32 s0, v40, 2
; GFX11-NEXT:    s_or_saveexec_b32 s1, -1
; GFX11-NEXT:    scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
; GFX11-NEXT:    s_mov_b32 exec_lo, s1
; GFX11-NEXT:    s_mov_b32 s33, s0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  call void @external_void_func_bf16_inreg(bfloat inreg %arg)
  ret void
}

define void @test_call_external_void_func_v2bf16_inreg(<2 x bfloat> inreg %arg) #0 {
; GFX9-LABEL: test_call_external_void_func_v2bf16_inreg:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    s_mov_b32 s17, s33
; GFX9-NEXT:    s_mov_b32 s33, s32
; GFX9-NEXT:    s_or_saveexec_b64 s[18:19], -1
; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
; GFX9-NEXT:    s_mov_b64 exec, s[18:19]
; GFX9-NEXT:    v_writelane_b32 v40, s17, 2
; GFX9-NEXT:    s_addk_i32 s32, 0x400
; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
; GFX9-NEXT:    s_getpc_b64 s[18:19]
; GFX9-NEXT:    s_add_u32 s18, s18, external_void_func_v2bf16_inreg@rel32@lo+4
; GFX9-NEXT:    s_addc_u32 s19, s19, external_void_func_v2bf16_inreg@rel32@hi+12
; GFX9-NEXT:    s_mov_b32 s0, s16
; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
; GFX9-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; GFX9-NEXT:    v_readlane_b32 s31, v40, 1
; GFX9-NEXT:    v_readlane_b32 s30, v40, 0
; GFX9-NEXT:    s_mov_b32 s32, s33
; GFX9-NEXT:    v_readlane_b32 s4, v40, 2
; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
; GFX9-NEXT:    s_mov_b32 s33, s4
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: test_call_external_void_func_v2bf16_inreg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s1, s33
; GFX11-NEXT:    s_mov_b32 s33, s32
; GFX11-NEXT:    s_or_saveexec_b32 s2, -1
; GFX11-NEXT:    scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
; GFX11-NEXT:    s_mov_b32 exec_lo, s2
; GFX11-NEXT:    v_writelane_b32 v40, s1, 2
; GFX11-NEXT:    s_add_i32 s32, s32, 16
; GFX11-NEXT:    s_getpc_b64 s[2:3]
; GFX11-NEXT:    s_add_u32 s2, s2, external_void_func_v2bf16_inreg@rel32@lo+4
; GFX11-NEXT:    s_addc_u32 s3, s3, external_void_func_v2bf16_inreg@rel32@hi+12
; GFX11-NEXT:    v_writelane_b32 v40, s30, 0
; GFX11-NEXT:    v_writelane_b32 v40, s31, 1
; GFX11-NEXT:    s_swappc_b64 s[30:31], s[2:3]
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_readlane_b32 s31, v40, 1
; GFX11-NEXT:    v_readlane_b32 s30, v40, 0
; GFX11-NEXT:    s_mov_b32 s32, s33
; GFX11-NEXT:    v_readlane_b32 s0, v40, 2
; GFX11-NEXT:    s_or_saveexec_b32 s1, -1
; GFX11-NEXT:    scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
; GFX11-NEXT:    s_mov_b32 exec_lo, s1
; GFX11-NEXT:    s_mov_b32 s33, s0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  call void @external_void_func_v2bf16_inreg(<2 x bfloat> inreg %arg)
  ret void
}